Professional Cloud Architect deep review — September 29, 2026

Same-context AI review; independent human review pending. Study guide.

Scope and lifecycle

The actual seven-page standard exam PDF and all twelve case-study pages were read with layout extraction. The mapping has 101 statements: 96 bullets and five standalone section-six objectives under 22 numbered objectives. Groups are 33/23/13/13/11/8; weights remain 25/17.5/17.5/15/12.5/12.5. Actual bytes, hashes and page counts are retained; no printed revision date is invented.

The objective digest remains fceaa0587ed0fa26fcf66a037a2abb7e4103231cccb6ad9052153ca3cb328ec2. A missing lifecycle baseline was explicitly initialized after reviewing the canonical page, at e914ad2789c7bdf089beb9e0650c2e050f74e7d75e60820cf55a146896fa5d03. A subsequent observation is unchanged. Before/accepted/post receipts distinguish initialization from a changed exam. Standard, shorter renewal-exam and designated-learning renewal routes are separated; renewal-specific PDFs were not separately reviewed.

Cases and teaching

The four public cases now have distinct facts, original design hypotheses and proposed evidence. Altostrat's existing media/hybrid estate and moderation needs are preserved. Cymbal's human review before catalog publication becomes an exact-content/version requirement. EHR retains its current insurance interfaces; its 99.9% target does not supply an RTO. KnightMotives' dealer-equipment budget is separated from vehicle modernization, and its fictional safety assertion is not treated as independently verified performance. No table is an official answer key.

The guide adds recovery dependency paths and migration catch-up arithmetic, regional HA versus asynchronous DR, ordinary promotion versus edition-specific advanced recovery, old-primary fencing and restored redundancy. DR planning guidance supports testing identity, artifacts, capacity and end-to-end recovery; historical compliance-version/marketing statements in that article are not adopted as current standards.

Private Service Connect feature direction, VPC Service Controls rehearsal versus enforcement, direct KMS rotation versus resource-specific CMEK behavior and attestation versus current vulnerability/admission evidence gain explicit boundaries. Previously read same-session primary sources supply Model Armor integration/modality, retention, retrieval identity, Registry and ADC/GKE distinctions. The April 22, 2026 app/platform announcement supplies naming context only.

The Google SRE alerting chapter was read selectively for burn rates, long/short-window confirmation and low-traffic limitations. One source example labels one failed request among ten at a 99.9% SLO as 1,000×. The formula gives 100×, independently calculated and tested; both cached text and the live web passage were checked. No production alert was deployed.

Executed local evidence

The exact public Python worksheet passed 32 checks: nine deterministic recovery-graph cases, seven steady-rate backlog cases, six SLO/arithmetic cases and ten static approval-binding cases. Public code hash and Python version are recorded. A fictional plan yields 68 minutes, versus 76 minutes if every task is incorrectly serialized; a shorter database task yields 53 minutes. A 900 GiB backlog with 100 MiB/s effective transfer and 80 MiB/s incoming writes takes 12.8 hours, versus 2.56 hours without new writes. Equal or higher arrivals do not converge.

The approval model rejects changed/stale content, wrong decision, missing records and untrusted reviewers under explicitly trusted input assumptions. Hashing does not authenticate the reviewer, prove human review/content correctness or implement atomic publication. Recovery duration and throughput are invented assumptions, not measurements; no cloud, clinical or vehicle system was tested.

There are 48 answered checks, three integrated scenarios, four official-case analyses and eight proposed cloud labs with acceptance/failure/recovery evidence. The prior ACE observation that gcloud is absent from PATH is reused; no installation, authentication, project, migration, restore, perimeter, signer, model or notification was attempted.

Learning resources and verification

Google Skills confirms 24 activities but not the earlier 172h15m/72h durations. Coursera now lists seven cards totaling 48 hours, including Gemini Notebook study preparation; landing and FAQ schedules disagree. A current Pluralsight path has seven courses totaling 7h17m plus a 30-minute August 6, 2026 lab, or 7h47m versus the rounded eight-hour header. Its course dates are July–December 2025. Titles and dates establish public catalog facts, not current paid-lesson coverage.

Whizlabs exposed a title-only shell and O'Reilly returned 403. Prior book metadata is qualified as unverified. No paid lessons, provider labs, proprietary assessments or recalled exam questions were read. Places to learn is last.

The operational receipt records local execution and, once completed, unit tests, repository validation, catalog consistency, strict site generation, generated-site validation and diff checks. Source freshness is current; the review outcome remains reviewed with blockers for live-cloud and independent-human verification.